R21 Night Dyads

Exploratory

List of previous requests

• Consider proximity based on Bluetooth pings and Morning Sleep Diary
  
• Calculate a measure of sleep-wake concordance that incorporates the continuous activity/movement data
  
• Calculate time spent in different sleep-wake states
  
• Exploratory analysis of human animal bond and veteran health
  
• Characterize sleep-wake transmission
  

Sample description

• ClinicalTrials.gov ID: NCT03245814
• K9FW R21
• 56 veterans paired with service dogs, based on available actigraphy data

Exclusions

Epoch-level exclusions

Epoch-level exclusions	Freq	Cumulative Freq	Percent	Cumulative Percent
1. Not in the sleep period	1364837	1364837	80.19	80.19
2. Missing veteran activity	547	1365384	0.03	80.22
3. Missing dog activity	35436	1400820	2.08	82.30
4. Missing Purdue sleep/wake classification	15	1400835	0.00	82.30
NA	301245	1702080	17.70	100.00

Night-level exclusions

Night-level exclusions	Freq	Cumulative Freq	Percent	Cumulative Percent
1. No veteran activity during the sleep period	1	1	0.15	0.15
2. No dog movement during the sleep period	25	26	3.70	3.85
NA	650	676	96.15	100.00

Proximity during the sleep period

Based on Bluetooth ping within prior 15 minutes

Minutes per night

Percent of the sleep period

Based on Morning Sleep Diary

Minutes per night with service dog in the room

Percent of the sleep period spent with service dog in the room

Minutes per night with service dog in the bed

Percent of the sleep period spent with service dog in the bed

Compare proximity based on Bluetooth with morning sleep diary

Minutes per night

Percent of the sleep period

Summary Tables

Minutes per night

Characteristic	N = 56
Bluetooth proximity, mean nightly minutes
Range	175, 543
Median (IQR)	396 (342, 443)
Mean (SD)	386 (81)
Service dog in the room, mean nightly minutes
Range	0, 579
Median (IQR)	393 (319, 465)
Mean (SD)	372 (135)
Missing	5
Service dog in the bed, mean nightly minutes
Range	0, 535
Median (IQR)	327 (86, 438)
Mean (SD)	274 (189)
Missing	5

Percent of the sleep period

Characteristic	N = 56
Bluetooth proximity, mean nightly percent
Range	38.4, 99.8
Median (IQR)	90.7 (80.6, 98.2)
Mean (SD)	86.8 (14.1)
Service dog in the room, mean nightly percent
Range	0.0, 100.0
Median (IQR)	95.5 (75.4, 100.0)
Mean (SD)	82.4 (27.2)
Missing	5
Service dog in the bed, mean nightly percent
Range	0.0, 100.0
Median (IQR)	81.8 (20.1, 100.0)
Mean (SD)	60.9 (40.3)
Missing	5

Sleep-wake concordance

Percent agreement

Cohen’s kappa

Kendall’s Tau

We want to characterize the extent to which the veteran and service dog were active at the same times during the night (without dichotomizing the activity data).

Challenges:

• Different measures of activity for veterans and service dogs, with different scales/units
  
  • Veteran activity was collected with the Actiwatch-2/Actiware software.
    
  • Service dog activity was collected with the Whistle.
    
• Zero-inflated activity data: Activity counts during the night are mostly zeros.
  
  • XX% of veteran epochs have an activity count equal to zero.
    
  • YY% of service dog epochs have an activity count equal to zero.
    

We can approach this by ranking the epoch-level activity data for the veteran, ranking the epoch-level activity data for the service dog, and then compare the ordering of the rankings each night to get a nightly measure of the extent to which the veteran and service dog were active during the same epochs.

Kendall’s tau is a measure of rank correlation. There are different versions of Kendall’s tau that handle ties in different ways. For most of the epochs the ranks are the same (i.e. due to zeros) for both the veteran and the dog, representing ties that need to be handled.

Pimentel, et al proposed an estimator based on Kendall’s tau for zero-inflated continuous variables. Perrone et al provided R code in Supplementary materials to calculate the estimator proposed by Pimentel et al, as well as their proposed estimator of Kendall’s tau for bivariate zero-inflated count data.

References:

• Pimentel, R. S., Niewiadomska-Bugaj, M., & Wang, J. C. (2015). Association of zero-inflated continuous variables. Statistics & Probability Letters, 96, 61-67.
  
• Perrone, E., van den Heuvel, E. R., & Zhan, Z. (2023). Kendall’s tau estimator for bivariate zero-inflated count data. Statistics & Probability Letters, 199, 109858.

How are these measures of sleep-wake concordance related?

Summary Table

Characteristic	N = 56
Mean nightly percent agreement
Range	0.63, 0.88
Median (IQR)	0.80 (0.76, 0.83)
Mean (SD)	0.79 (0.05)
Mean nightly Cohen's kappa
Range	0.31, 0.85
Median (IQR)	0.73 (0.66, 0.78)
Mean (SD)	0.71 (0.10)
Mean nightly Kendall's Tau
Range	-0.02, 0.15
Median (IQR)	0.04 (0.03, 0.06)
Mean (SD)	0.04 (0.03)

Sleep-wake states - Summary Tables

Minutes per night

Characteristic	N = 56
Concordant - both sleeping, mean nightly minutes
Range	190, 449
Median (IQR)	330 (302, 368)
Mean (SD)	334 (60)
Concordant - both awake, mean nightly minutes
Range	5, 83
Median (IQR)	16 (12, 22)
Mean (SD)	18 (11)
Discordant - veteran sleep, dog awake, mean nightly minutes
Range	19, 89
Median (IQR)	46 (38, 56)
Mean (SD)	48 (16)
Discordant - veteran awake, dog sleeping, mean nightly minutes
Range	6, 115
Median (IQR)	43 (31, 60)
Mean (SD)	46 (22)

Percent of the sleep period

Characteristic	N = 56
Concordant - both sleeping, mean nightly percent
Range	45.4, 84.8
Median (IQR)	76.2 (70.9, 79.8)
Mean (SD)	74.8 (7.2)
Concordant - both awake, mean nightly percent
Range	1.2, 17.4
Median (IQR)	3.5 (2.6, 4.6)
Mean (SD)	3.9 (2.4)
Discordant - veteran sleep, dog awake, mean nightly percent
Range	4.2, 24.9
Median (IQR)	10.4 (9.0, 12.1)
Mean (SD)	10.8 (3.8)
Discordant - veteran awake, dog sleeping, mean nightly percent
Range	1.9, 26.4
Median (IQR)	9.8 (6.5, 12.7)
Mean (SD)	10.5 (5.2)

Exploratory analysis of human-animal bond and veteran health

Measures of veteran-canine bond

Inclusion of other in self

Monasch Dog Owner Relationship Scale

Lexington Attachment to Pets Scale

Measures of veteran health

PCL-5 Total Score

Fear of Sleep Inventory (FOSI)

Wake Transmission

Walters et al calculated four transmission variables.
1. Percent transmissions received: the percentage of an individual’s awakenings in a given time interval classified as a wake transmission (i.e., total number of transmitted wakes received by individual/total awakenings in individual).
2. Percent minutes resistant to transmission: percentage of a bedpartner’s nocturnal wake minutes during which an individual slept.
3. Number of transmissions received: the total number of transmission events in a given time interval.
4. Transmissibility: the percentage of the bedpartner’s awakenings in a given time interval that were classified as wake transmissions (i.e., number of wake transmissions received/total bedpartner awakenings: a measure of vulnerability to a bedpartner’s awakenings.

Reference:
- Walters, E. M., Phillips, A. J., Boardman, J. M., Norton, P. J., & Drummond, S. P. (2020). Vulnerability and resistance to sleep disruption by a partner: A study of bed-sharing couples. Sleep health, 6(4), 506-512.

Characteristic	N = 56
Percent transmissions received - veteran
Range	6.6, 32.7
Median (IQR)	13.8 (11.3, 18.6)
Mean (SD)	15.1 (5.7)
Percent transmissions received - service dog
Range	3.9, 49.0
Median (IQR)	22.2 (15.2, 25.8)
Mean (SD)	21.5 (9.2)
Percent minutes resistant to transmission - veteran
Range	40.3, 91.2
Median (IQR)	74.3 (68.6, 80.0)
Mean (SD)	73.3 (10.2)
Percent minutes resistant to transmission - service dog
Range	43.2, 87.4
Median (IQR)	72.1 (65.9, 77.5)
Mean (SD)	71.2 (9.2)
Number of transmissions received - veteran
Range	1.2, 6.8
Median (IQR)	3.3 (2.5, 4.4)
Mean (SD)	3.5 (1.4)
Number of transmissions received - service dog
Range	1.1, 15.1
Median (IQR)	6.0 (4.4, 8.5)
Mean (SD)	6.7 (3.2)
Transmissibility - veteran
Range	4.2, 20.9
Median (IQR)	10.1 (8.8, 13.5)
Mean (SD)	11.1 (3.5)
Transmissibility - service dog
Range	8.6, 82.1
Median (IQR)	25.5 (20.0, 32.5)
Mean (SD)	27.9 (13.0)